What downtime and production bugs really cost a business, and how much to spend on prevention
In July 2024 one faulty update from CrowdStrike crashed about 8.5 million Windows computers. The insurer Parametrix estimated the direct losses for US Fortune 500 companies alone at $5.4 billion. Few companies will ever have an outage that size, but every business that sells through software has a price for an hour of downtime, and few founders know theirs.
Below we collect what recent research says about outage costs, explain why those numbers rarely fit a small business, and show how we size the spend on monitoring, staged releases and backups against the actual risk.
Three numbers set the scene:
$2M median cost of an hour of high-impact outage (New Relic, 2025)
57% of major outages cost more than $100,000 (Uptime Institute, 2026 report)
$5.4B estimated Fortune 500 loss from one bad update (Parametrix on CrowdStrike, 2024)
What the surveys say
Four studies are quoted most often. They ask different people different questions, so their numbers should not be compared directly.
ITIC, 2024 (1,000+ firms worldwide): an hour of downtime costs over $300,000 for more than 90% of mid-size and large firms.
Splunk and Oxford Economics, 2026 (2,000 executives at Global 2000 companies): $600 billion a year in total, $15,000 per minute on average.
New Relic Observability Forecast, 2025 (1,700 IT and engineering leaders in 23 countries): a median of $2 million per hour of high-impact outage.
Uptime Institute, 2026 (operators in its 2025 annual survey): 57% say their last major outage cost over $100,000, and 1 in 5 over $1 million.
All four are estimates reported by the respondents themselves, and Splunk, New Relic and Uptime Institute sell products or services related to reliability.
Read them as a direction, not as a price tag for your company. They describe large enterprises, and most of the publishers sell monitoring. Two findings still carry over to a company of any size. First, outages get rarer but not cheaper: Uptime reports falling outage frequency per site for five years running while costs keep rising. Second, the obvious loss is rarely the biggest one.
One update, 8.5 million machines
CrowdStrike’s own root cause analysis is short and worth reading. A new sensor feature had shipped in February 2024 and several content updates for it worked fine. On 19 July an update supplied 21 input fields where the sensor expected 20. The sensor read past the end of its data and crashed the machine it was protecting.
How it unfolded in 2024:
5 March: the first content for the new sensor capability ships without problems.
19 July: an update with 21 fields where the sensor expects 20. About 8.5 million Windows devices crash (Microsoft’s estimate).
24 July: Parametrix estimates a $5.4 billion loss for the Fortune 500, of which only 10% to 20% is likely to be insured.
29 July: about 99% of sensors are back online, ten days later.
6 August: CrowdStrike’s root cause analysis lists the fixes: more tests, deployment rings and customer control over updates.
Parametrix put the weighted average loss at $44 million per affected Fortune 500 company, from $6 million in manufacturing to $143 million for airlines. The lesson for a small team is in the fix list, not the size: automated tests for every template type, updates that pass successive deployment rings before reaching everyone, and customers who can choose when updates arrive. None of that is exotic. It is the routine work that gets postponed because the last fifty releases were fine.
The costs that never reach an invoice
Lost sales during the outage are the part everyone counts. The rest is larger and slower.
Customers. In Splunk’s 2026 study, 81% of technology leaders name losing customers as a consequence of downtime, and 47% say customers are often or very often the first to notice an outage. Public companies saw an average 3.4% share price drop after a downtime event.
Engineering time. New Relic’s respondents say engineers spend 33% of their time fighting fires. That is roadmap you paid for and did not get.
App store reach. Google Play may reduce an app’s visibility and show a warning on its listing when the user-perceived crash rate goes above 1.09% overall or 8% on any phone model. Play judges a 28-day average, so one bad release can cost you for a month.
Slowness. A slow app is a partial outage. In a Google-commissioned study by 55 and Deloitte of 37 brands (data from late 2019), a 0.1 second faster mobile site went with 8.4% more retail conversions. It is a correlation and the data is older, but the direction is consistent.
You will often see the claim that a bug costs 100 times more to fix in production. We do not use it: Laurent Bossavit traced the usual “IBM Systems Sciences Institute” source to internal training course notes with no published data. The difference is real anyway, and you can see it without a multiplier.
A bug found before release means one developer fixes it, there is one more code review, and nobody outside the team notices. The same bug found by customers means hours before anyone knows, a hotfix plus an app store review, support tickets and refunds, bad data to repair, and reviews that stay for months.
How to size the investment
The research gives you direction. Your own arithmetic gives you a budget. The formula is simple: incidents per year, times hours to detect and recover, times what an hour costs you. Here is a worked example with assumptions we made up for illustration.
A company sells $2 million a year through its app, about $228 per hour on average. Incidents tend to show up when traffic is high, so we take twice the average, about $460 an hour, and add $225 an hour for three engineers working on the fix. That is roughly $685 for every hour of an incident. Assume six incidents a year.
In our calculation, if customers are the ones who report each problem and it takes 4 hours to detect and fix, the six incidents cost about $16,440 a year. With alerts and a one-click rollback bringing each incident down to 45 minutes, they cost about $3,083. These are our own numbers from the assumptions above, not survey data.
The gap, about $13,000 a year, is the direct budget the basics have to fit in, and at this size they usually do: an error tracker, uptime checks and a rollback path take days to set up, not months. The calculation leaves out the expensive tail: a lost database, a payments bug, a store warning in launch week. That tail is why backups and restore drills come first, even when the arithmetic looks small.
What we put in place, by stage:
First users: error tracking, uptime checks and tested daily backups, so you hear about problems before customers do and data loss is recoverable.
Paying customers: alerts to a named person, one-click rollback and staged rollouts in both app stores, so hours of recovery become minutes.
Revenue depends on uptime: an on-call rota, a written incident process and short post-incident reviews, so the same incident does not happen twice.
Contracts with SLAs: redundancy, load tests and regular restore drills, so the SLA penalty stops being a risk you carry.
What to do this month
Work out your cost per hour. Revenue per busy hour plus the cost of the people who fix things. Write it down where the team can see it.
Count last year’s incidents. How many, and how long until someone noticed. Detection time is usually the biggest number.
Restore a backup. A backup you have never restored is a hope, not a plan.
Release in stages. Google Play staged rollouts and App Store phased releases cost nothing and limit the damage of a bad build.
If you want a second pair of eyes on your release process or monitoring, talk to us: https://www.commonsaltx.com/contact
